PyTorch explains compile-time heuristics that pick Helion kernel configs without autotuning
AIPyTorch says Helion now uses compile-time heuristics that analyze a kernel's IR and dataflow to choose block sizes, warps, and stages without autotuning. Kernels are classified as matmul, reduction, or pointwise, and a family-specific heuristic emits the config.
Source details
PyTorch Blog · Caleb Min Sup Kim, Ethan Che, Jongsok Choi, Jason Ansel · pytorch.orgPublished
Original title: Fast by Default: Compile-Time Heuristics for Helion Kernels